==========================
MAJOR FEATURES + CHANGES
==========================
Change 2816560 on 2016/01/05 by Jeff.Campeau
Remove duplicate CEF binaries
Change 2835599 on 2016/01/20 by Lee.Clark
PS4 - Added pragma optimization macros
Change 2841103 on 2016/01/23 by Mark.Satterthwaite
Integrate Git PR #1958: Fixed typo in EMetalFeatures enum
#jira UE-25721
Change 2841369 on 2016/01/24 by Mark.Satterthwaite
Fix for Metal crash due to attempt to set a null uniform & null SRV to a shader which requires both exist.
#jira UE-25910
Change 2841795 on 2016/01/25 by Lee.Clark
PS4 - MovieStreamer improvements
* Use GPU for YUV conversion
* Use new Software2 Decoder
Change 2842261 on 2016/01/25 by Mark.Satterthwaite
Fix some memory leaks.
Change 2842831 on 2016/01/25 by Mark.Satterthwaite
Metal implementation for RHIBlockUntilGPUIdle.
Change 2842838 on 2016/01/25 by Mark.Satterthwaite
When using parallel command contexts in Metal we must ensure that the FRingBuffer is still valid, which means some smart/weak pointers are in order. We should also ensure that functions that may return auto-released objects are appropriately wrapped with scoped autorelease pools. Texture creation failures should also be fatal as we never expect that to occur.
Change 2842914 on 2016/01/25 by Mark.Satterthwaite
Change assert in MetalTexture's format-shifting SRV constructor to enforce the Metal textureView limits:
1. No format shifting for MSAA color buffers.
2. No access to stencil in MSAA packed depth/stencil surface.
This will allow Metal MSAA support to work on iOS when using separate depth & stencil textures since there's no format shifting involved there.
#codereview peter.sauerbrei
Change 2843028 on 2016/01/25 by Mark.Satterthwaite
In Metal wwitch to blit on the correct context when copying out the stencil data into the stencil SRV copy.
Change 2845531 on 2016/01/27 by Lee.Clark
PS4 - Fix memory alignment for back buffers
* Fix memory alignment for MapLargeBlock
* Fix available direct memory tracking
Change 2846491 on 2016/01/27 by Jeff.Campeau
2015 compile fixes for Orion
Change 2847395 on 2016/01/28 by Mark.Satterthwaite
Clear the stencil-SRV copy to 0 in Metal using a blit when created to avoid artefacts if used prior to the parent texture being rendered.
#jira UE-25834
Change 2847419 on 2016/01/28 by Mark.Satterthwaite
Apply the same fix to OpenGL's Stencil SRV logic as CL #2847395 applies to Metal.
Change 2848093 on 2016/01/28 by Mark.Satterthwaite
Cache parallel encoding Metal contexts & reuse them rather than creating a new one each time in order to massively improve parallel encoding performance. This required adding a reset function to Metal's internal state-cache which calls the CommandEncoder wrapper's reset so we don't accidently retain previous state.
Change 2849469 on 2016/01/29 by Mark.Satterthwaite
Defer render & compute command encoder construction to draw/dispatch etc to eliminate redundant encoders that then perform unnecessary driver & GPU synchronisation work. Currently Clear loadActions force an encoder even if it would then be empty as otherwise we see incorrect rendering. This needs to be tracked and optimised away too in order to achieve the same performance as D3D11.
Change 2849820 on 2016/01/29 by Daniel.Lamb
Fixed issue where a single DDC back end would not create a hierarchy.
#codereview Peter.Sauerbrei
Change 2850762 on 2016/02/01 by Jeff.Campeau
System-wide critical section support for Xbox One
Change 2850763 on 2016/02/01 by Jeff.Campeau
Network and product config for Orion
Change 2852459 on 2016/02/02 by Mark.Satterthwaite
Temporarily disable the lazy render command-encoder construction while investigating why it turns some samples black in Metal SM5 mode.
Change 2853947 on 2016/02/03 by Mark.Satterthwaite
Fix some lazy encoder construction fallout which also means we don't need to recreate render encoder state when performing profiling - the next draw/clear will do that as required.
Change 2854015 on 2016/02/03 by Mark.Satterthwaite
Move Stencil SRV blitting into FMetalSurface::UpdateSRV called when binding the texture SRV instead of having it done immediately post-rendering. This should avoid paying for the blit when stencil SRV sampling is never used or multiple blits when render-encoders that write stencil are split up due to query buffer overflow or similar. The cost will be a blit per-bind instead which should be more predictable.
Change 2854142 on 2016/02/03 by Mark.Satterthwaite
Implemented GetTextureBaseRHI (brought over from Dev-Rendering CL #2853948) for Metal to avoid unnecessary virtual function call chain to resolve the FMetalSurface* from an RHI texture.
Change 2854222 on 2016/02/03 by Mark.Satterthwaite
Remove the uniform buffer resource caching from Metal to match Dev-Rendering CL #2853948.
Change 2854246 on 2016/02/03 by Mark.Satterthwaite
Removed the uniform buffer resource caching from OpenGLDrv & implemented GetTextureBaseRHI to avoid unnecessary virtual function calls to match Dev-Rendering CL #2853948.
Change 2854279 on 2016/02/03 by Mark.Satterthwaite
Remove direct access to the MTLCommandQueue, for parallel rendering to work we're going to need to do a bit of management that means its more sensible to keep it private.
Change 2855524 on 2016/02/04 by Lee.Clark
PS4 - Fix Grayscale SRGB support
[CL 2898161 by Josh Adams in Main branch]
* Modules that use #import (DatabaseSupport & VisualStudioSourceCodeAccessor) are forced to be built locally due to SN-DBS limitation . These are very small modules so have a minimal impact on build time.
#codereview Mike.Fricker
[CL 2610243 by Lee Clark in Main branch]
- Two actual bugs were fixed (GitSourceControlUtils.cpp, VisualStudioSourceCodeAccessor.h)
- (These errors were found by using the Clang compiler on Windows platform)
#codereview marc.audy
[CL 2532216 by Mike Fricker in Main branch]
We already had FPlatformProcess::CreateProc/CloseProc pair (which is part of larger set with WaitForProc(), TerminateProc(), GetProcReturnCode(), etc). FProcHandle::Close() made the API more confusing (and actually introduced bugs, because it did not call CloseProc() until recently).
#codereview Josh.Adams, Robert.Manuszewski, Jaroslaw.Surowiec
[CL 2476322 by Dmitry Rekman in Main branch]
We now save the project when we add new files to via the editor. This avoids VS showing a prompt about the .vcxproj being dirty when closed.
[CL 2462427 by Jamie Dale in Main branch]
Stopped the "Compiling C++ Code" message in the editor MainFrame always having a cancel button
FModuleCompilerStartedEvent now passes a bool to say whether the hot-reload/compile operation is being run async, or whether it's blocking. We now only show the cancel button for async compiles, as the UI is otherwise blocked waiting for the compile to finish.
Added a bWaitForCompletion input parameter to DoHotReloadFromEditor. This mirrors the RebindPackages API that DoHotReloadFromEditor internally calls, and prevents UPackFactory needing to perform a manual tick.
[CL 2437789 by Matthew Griffin in Main branch]
Added a way to use DTE to add new files directly into VS without having to generate projects
This avoids the annoying "Do you want to reload?" dialog that appears in VS when projects are externally changed, and also saves us the cost/time of having to run UBT to generate projects.
[CL 2435515 by Matthew Griffin in Main branch]
Allowed the DTE VS integration to retry a file open operation if it detected a blocked instance of VS
UE-8623 - Adding new code to project opens new Visual Studio rather than using existing one
We'd previously disabled the defered opening (essentially a retry) of files if VS detection failed as it could get stuck in a loop when using a non-DTE version of VS (VS Express), however there are timing issues when we add new classes in the editor where an existing VS instance can be blocked waiting on a modal dialog.
I've changed our VS integration to report these DTE based VS instances as "blocked" rather than "unknown", and we now allow a defer on these until they become unblocked.
ReviewedBy Thomas.Sarkanen
[CL 2433390 by Matthew Griffin in Main branch]
There were two problems with this. One, I had to suppress registering build platforms in UBT if the project has no code. To achieve this I added new flag -nocodeproject and I'm passing it down from the editor to UAT.
The second one emerged after fixing previous one. The Editor was adding -build flag to the UAT if it found any of the VS 2012 or 2013. -build flag is making UAT to try to build, which fails as it is not supported on VS 2012-only system. During the discussion with platform team we came up with the solution to not check for VS 2012 in editor, as we only support it for XDK development, which requires both VSes on the system (2012 and 2013). Hence flag -build should be appended to the UAT command line if and only if there is VS2013 on the system.
[CL 2431415 by Jaroslaw Palczynski in Main branch]
* Multicast delegate Add* calls now return FDelegateHandles, and Remove* calls are now all deprecated, except for a new Remove function which takes a FDelegateHandle.
* New FConsoleManager::RegisterConsoleVariableSink_Handle and UnregisterConsoleVariableSink_Handle functions which work in terms of FConsoleVariableSinkHandle.
* Timer calls which don't take FTimerHandles are deprecated.
* FTicker::AddTicker now returns an FDelegateHandle and is removed by an overloaded Remove function.
* DEFINE_ONLINE_DELEGATE* macros now define _Handle variants of the Add/Remove functions which return/take handles.
* Various other handle-based registration changes.
* Some unity build fixes.
* Some simplification of delegate code.
* Fixes for lots of existing code to use handle-based registration and unregistration.
#codereview robert.manuszewski
[CL 2400883 by Steve Robb in Main branch]
Made sure to manually load the appropriate modules for source code access.
The hot reload dependency that VisualStudioSourceCodeAccess had was bringing in the whole editor as a dependency, so this is now restricted to editor builds.
SlateViewer now compiles in CoreUObject as the source code accessor module requires it for its settings.
TTP# 333675 - CHECKIN: SLATE: WidgetReflector code reference hyperlinks don't work anymore
#codereview Nick.Atamas,Frank.Fella
[CL 2292278 by Thomas Sarkanen in Main branch]
TTP# 331398 - EDITOR: Logspam when opening Visual Studio
Deferring a VS file open request used to use a deferred command to wait until VS had finished opening before opening the requested files, however, this command was essentially simulating a Tick, as if the command was not yet ready to be processed it was simply deferred again, leading to spammed output.
Rather than do this, I've added a Tick function which will attempt to open the entire list of pending files as a single call, rather than try each file individually like the command would.
ReviewedBy Thomas.Sarkanen, Chris.Wood, Max.Preussner
[CL 2058815 by Jamie Dale in Main branch]
